diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-10-18 10:21:02 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:55:45 +0000 |
| commit | 4b90865ab985d571f70c60583cdfb8c7a65f1670 (patch) | |
| tree | f116a4ffef5f5e823689dd00c1e5c9d987f3d295 /src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp | |
| parent | c55beee7ef70fa08a5d217619083b288a74fcb27 (diff) | |
| download | ComputeLibrary-4b90865ab985d571f70c60583cdfb8c7a65f1670.tar.gz | |
COMPMID-1413 - Improve the performance of GEMMLowp with 8 bit dot product on OpenCL
COMPMID-1424 - Add dot product support for CLDepthwise QASYMM8 3x3 NHWC non-unit stride
With this patch we are able to improve the performance of MobileNet v1-qasymm8 by 37 %
Tried to use the dot product instruction in CLDepthwise QASYMM8 3x3 NHWC non-unit stride
but I have not seen any benefit (maybe because we have few arithemtic operation and we
do not have more load instructions). However Depthwise convolution has been improved by
30%
Change-Id: Id768a99c2e53a04276707e427af5d0ec93419ada
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/155082
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp | 3 |
1 files changed, 1 insertions, 2 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp index eb561faf77..19cc649c96 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp @@ -246,13 +246,12 @@ void CLDepthwiseConvolutionLayer3x3NCHWKernel::configure(const ICLTensor *input, int output_shift = 0; quantization::calculate_quantized_multiplier_less_than_one(multiplier, &output_multiplier, &output_shift); + build_opts.add_option("-DREAL_MULTIPLIER=" + support::cpp11::to_string(multiplier)); build_opts.add_option("-DCONV_STRIDE_Y=" + support::cpp11::to_string(_conv_stride_y)); build_opts.add_option("-DINPUT_OFFSET=" + support::cpp11::to_string(-_input->info()->quantization_info().offset)); build_opts.add_option("-DWEIGHTS_OFFSET=" + support::cpp11::to_string(-_weights->info()->quantization_info().offset)); build_opts.add_option("-DOUTPUT_OFFSET=" + support::cpp11::to_string(_output->info()->quantization_info().offset)); build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(9 * input->info()->quantization_info().offset * weights->info()->quantization_info().offset)); - build_opts.add_option("-DOUTPUT_MULTIPLIER=" + support::cpp11::to_string(output_multiplier)); - build_opts.add_option("-DOUTPUT_SHIFT=" + support::cpp11::to_string(output_shift)); if(act_info.enabled()) { |